The Invariant Structures Beneath Change.

In December 1930, depositors gathered outside the Bank of United States in New York and demanded their money. In March 2023, Silicon Valley Bank depositors did not need a street, a teller, or even a banking day. Messages moved through private chats and social feeds; withdrawals moved through screens. By the end of March 9, $42 billion had left the bank.

The images belong to different worlds. One is a black-and-white queue on a sidewalk. The other is invisible: a cascade of messages, browser sessions, and wire instructions. Yet the underlying mechanism is almost identical. Depositors believe others will leave first. Rational self-protection becomes collective destruction. A stable institution can move toward an unstable equilibrium because everyone expects everyone else to run.

In the past decade, I have watched one technology wave after another be declared unprecedented. Blockchain, autonomous systems, the metaverse, generative AI, agents—each arrived with a vocabulary suggesting that the old rules no longer applied. Each time, the tools genuinely changed. Each time, the old rules kept operating quietly underneath.

What changed was the carrier, the speed, and the instruments. What did not change was the structure.

What actually changes

It would be a mistake to turn this observation into nostalgia. Technology does change economic behavior in structural ways. Computation and communication improved inventory management. Digital distribution drove the marginal cost of many products toward zero. Cloud infrastructure transformed capital expenditure into a service. Social platforms collapsed the distance between observation, interpretation, and action.

The Great Moderation is a useful example. U.S. output and inflation volatility fell sharply after the mid-1980s. Better monetary policy and smaller shocks mattered, but advances in information technology and business practice also improved the economy’s ability to manage inventories and absorb disturbances. The old inventory cycle did not vanish; its amplitude changed, and some volatility migrated into other parts of the system.

SVB demonstrates the same distinction. Social media did not invent the bank run, but it compressed coordination time. Mobile banking did not create maturity mismatch, but it made deposits radically more mobile. The mechanism described by bank-run models remained intact while its transmission accelerated from days to hours.

This is the first discipline of thinking about the future: separate the variables that change from the relationships that persist. Speed can change. Scale can change. Interfaces, institutions, and participation can change. But incentives, feedback loops, coordination problems, and the dependence of long-term promises on short-term funding tend to survive the upgrade.

Invariant one: cycles

In 1862, Clément Juglar argued that commercial crises were not isolated accidents but phases in a recurring rhythm of prosperity, crisis, and liquidation. The exact length was never the point. The deeper insight was that investment creates capacity, success encourages extrapolation, extrapolation produces excess, and excess eventually forces adjustment.

Every generation gives this rhythm new machinery. Railways required tracks, rolling stock, and land. Electrification required generators and grids. The internet required fiber, servers, and data centers. AI requires chips, power, cooling, networking, models, and software capable of converting all that infrastructure into useful work.

The current buildout is already visible in corporate accounts. Amazon reported that cash capital expenditure rose from $77.7 billion in 2024 to $128.3 billion in 2025, primarily reflecting technology infrastructure and capacity for AWS growth. This does not prove that AI is a bubble. It proves something more basic: a general-purpose technology becomes an investment cycle once belief is translated into physical capacity.

Carlota Perez offers a helpful template. A technological revolution begins with an installation period, when financial capital funds new infrastructure, experiments proliferate, and speculation outruns deployment. A correction separates durable capability from financial excess. Then comes deployment: the technology diffuses through the economy and becomes ordinary enough to be productive.

The dot-com collapse did not invalidate the internet. It cleared the conditions for cloud computing, mobile distribution, and digital business models to spread. The overbuilding was both economically painful and technologically useful. That duality is the signature of an installation cycle.

This pattern also explains why technological truth and investment truth can diverge. A technology can be inevitable while a particular financing structure, valuation, or market entrant is not. Railways changed commerce even when railway shareholders were ruined. Fiber laid during the internet boom became useful long after the companies that financed it disappeared. Correctly identifying the destination does not guarantee that every vehicle reaches it.

Periods of calm tempt people to declare cycles obsolete. The Great Moderation produced exactly that confidence. But reduced volatility in inventories and consumer prices did not abolish instability; leverage and risk accumulated in the financial system instead, then released in 2008. Cycles do not die. They migrate.

Invariant two: narratives

Robert Shiller describes economic narratives as contagious stories that shape decisions at scale. The most powerful are perennial. They become dormant, mutate to fit new evidence, and return in a form that feels native to the present.

Consider the fear that machines will take human work. It animated resistance to mechanized looms in the 1810s. In the 1930s, John Maynard Keynes gave “technological unemployment” a name. Automation anxiety returned in the 1960s strongly enough for President Lyndon Johnson to establish a national commission. Today the same story travels through model benchmarks, agent demonstrations, and forecasts of artificial general intelligence.

The occupations, machines, and predictions change. The narrative structure does not: a new capability appears, observers extrapolate its steepest curve, and a complex reorganization of tasks is compressed into a binary contest between human and machine.

The persistence of this story does not make it false. That is precisely why it survives. Strong narratives attach themselves to partial truth and then extend beyond the evidence. The internet really did change almost every industry, but “eyeballs” were not earnings. AI really is a general-purpose technology, but that fact alone cannot validate every product, company, or valuation built around it.

A narrative becomes hard to kill when its critics must concede that half of it is right. It offers a true observation, an emotionally resonant consequence, and a simple direction of travel. That combination coordinates attention long before the final economic outcome is knowable.

For investors and builders, narrative is not decoration added after the fundamentals. It is part of the transmission mechanism. It determines which facts become salient, which experiments attract resources, and which possible futures people can coordinate around.

Invariant three: liquidity

Liquidity is the variable that seems secondary until it becomes the only variable that matters. In ordinary conditions it looks like a pricing adjustment: a tighter spread, a lower funding cost, a deeper market. Under stress it becomes a survival constraint. A solvent institution can fail when it cannot meet today’s obligations; a strong technology can stall when capital cannot bridge the distance between infrastructure and adoption.

Charles Kindleberger and Hyman Minsky described a recurring pattern in financial manias. The process begins with displacement—a genuine change in technology, policy, or markets. Credit expands to finance the opportunity. Rising prices validate the story, draw in participants, and relax discipline. Eventually the structure depends not only on future success, but on continuously available funding.

Technology provides the story; credit provides the fuel. That is why real innovation and financial excess so often arrive together. The better the underlying technology, the easier it becomes to justify funding every adjacent claim.

The form of liquidity changes by era. In 1930, the fragile instrument was the bank deposit. In 2008, it included wholesale funding and repurchase markets. In 2023, uninsured deposits could be coordinated through social networks and withdrawn digitally at extraordinary speed. The instruments differ; the structure is a sudden stop in financing.

The same lens matters for AI. The most useful question is not whether aggregate spending is “too high.” It is how the buildout is funded, what cash flows are expected to support it, where maturity or concentration risks sit, and how long capital providers can wait for deployment economics to appear. Liquidity sets both the fuel and the time limit.

The interaction is the invariant

The deepest constant is not any single cycle, story, or funding instrument. It is their interaction. Cycle describes the pendulum of investment, capacity, and credit. Narrative describes how attention spreads and how possibility becomes collectively legible. Liquidity describes the fuel available to turn belief into action—and the time before that fuel runs out.

This became CNL: Cycle, Narrative, Liquidity, a framework for reading any market where capital and collective belief interact. It was born in crypto’s compressed cycles, where decades of market structure play out at high frequency—but it applies wherever the three forces meet. What each force measures, and how to read the alignment among them, is the subject of its own essay.

The point here is narrower and older: structural thinking does not predict the future by assuming that history repeats literally. It separates the interface, which changes constantly, from the relationships underneath, which change slowly if at all—and it treats any claim that those relationships have been abolished as the most expensive claim in finance.

That discipline matters most at the peak of novelty, when velocity is mistaken for permanence and skepticism is mistaken for blindness. Structural thinking allows both truths to coexist: the new technology may be profound, and the market around it may still be repeating an old pattern. Holding both ideas at once is not indecision. It is the beginning of judgment.

The future will continue to surprise us. Models will improve, interfaces will disappear, institutions will be redesigned, and markets will move faster. But novelty at the surface should not prevent structural memory underneath.

The only thing that reliably becomes obsolete is the belief that this time is different. Technology changes the costume. Structure writes the play.

Sources & further reading

The historical and market references in this essay draw on the FDIC’s account of the 2023 regional bank failures, the Federal Reserve’s work on the Great Moderation, and Amazon’s 2025 annual filing. The conceptual lineage includes Clément Juglar, Carlota Perez, Robert Shiller, Charles Kindleberger, Hyman Minsky, and the Diamond–Dybvig model of bank runs.